Do not add type tag size to the alloc_typed lowering for GC allocations (#54837)

Enzyme.jl hit an issue where, in a dynamically typed allocation of size
`GC_MAX_SZCLASS`, because we mistakenly added they type tag size to the
allocation, the runtime disagreed if this was a pool allocation or a big
allocation. Causing a crash in the GC

(cherry picked from commit ded0b28)

Author: gbaraldi

src/julia_internal.h

@@ -508,6 +508,8 @@ STATIC_INLINE uint8_t JL_CONST_FUNC jl_gc_szclass_align8(unsigned sz) JL_NOTSAFE
 #define GC_MAX_SZCLASS (2032-sizeof(void*))
 static_assert(ARRAY_CACHE_ALIGN_THRESHOLD > GC_MAX_SZCLASS, "");
 
+
+// Size does NOT include the type tag!!
 STATIC_INLINE jl_value_t *jl_gc_alloc_(jl_ptls_t ptls, size_t sz, void *ty)
 {
     jl_value_t *v;

src/llvm-final-gc-lowering.cpp

@@ -211,7 +211,7 @@ void FinalLowerGC::lowerGCAllocBytes(CallInst *target, Function &F)
         }
     } else {
         auto size = builder.CreateZExtOrTrunc(target->getArgOperand(1), T_size);
-        size = builder.CreateAdd(size, ConstantInt::get(T_size, sizeof(void*)));
+        // allocTypedFunc does not include the type tag in the allocation size!
         newI = builder.CreateCall(allocTypedFunc, { ptls, size, type });
         derefBytes = sizeof(void*);
     }

test/llvmpasses/final-lower-gc.ll

@@ -93,9 +93,8 @@ top:
   %pgcstack = call {}*** @julia.get_pgcstack()
   %ptls = call {}*** @julia.ptls_states()
   %ptls_i8 = bitcast {}*** %ptls to i8*
-; CHECK: %0 = add i64 %size, 8
-; TYPED: %v = call noalias nonnull align {{[0-9]+}} dereferenceable(8) {} addrspace(10)* @ijl_gc_alloc_typed(i8* %ptls_i8, i64 %0, i64 12341234)
-; OPAQUE: %v = call noalias nonnull align {{[0-9]+}} dereferenceable(8) ptr addrspace(10) @ijl_gc_alloc_typed(ptr %ptls_i8, i64 %0, i64 12341234)
+; TYPED: %v = call noalias nonnull align {{[0-9]+}} dereferenceable(8) {} addrspace(10)* @ijl_gc_alloc_typed(i8* %ptls_i8, i64 %size, i64 12341234)
+; OPAQUE: %v = call noalias nonnull align {{[0-9]+}} dereferenceable(8) ptr addrspace(10) @ijl_gc_alloc_typed(ptr %ptls_i8, i64 %size, i64 12341234)
   %v = call {} addrspace(10)* @julia.gc_alloc_bytes(i8* %ptls_i8, i64 %size, i64 12341234)
   %0 = bitcast {} addrspace(10)* %v to {} addrspace(10)* addrspace(10)*
   %1 = getelementptr {} addrspace(10)*, {} addrspace(10)* addrspace(10)* %0, i64 -1